Optimizing COVID-19 Interventions While Accounting for Economy and Education

Overview

Utilizing reinforcement learning and deep neural networks to optimize COVID-19 interventions. Under the umbrella of artificial intelligence, machine learning is a technique that allows a system to improve through experience without any explicit programs. One approach to machine learning used in this study is neural networks, in which the computer learns to finish a task by analyzing training samples. The other approach used is reinforcement learning which manipulates a programmed environment and discovers errors and rewards to output the optimal solution. This study created a deep neural network and used reinforcement learning to develop a model that could predict whether the number of COVID-19 cases would increase or decrease. Then using that information, the model predicted which actions would most effectively cause a decline in cases while keeping things like economy and education in mind for a better long-term effect. These models were made based on eight different Florida counties’ data including mobility, temperature, dates of government actions, etc. Based on this information, data exploration and feature engineering were conducted to add dimensions that would further the accuracy of the neural network. The reinforcement learning model’s actions consisted of a shutdown for about two months before reopening schools and allowing things to return to normal. Then interestingly, the model decided to keep schools operating in a hybrid model, with some students going back to school while others continue to study remotely. These models can be generalized to train agents for taking action against policies in areas beyond COVID-19. Using neural networks and reinforcement learning in government decision-making could push for better measurement and consideration for education, local economy, and other essential factors that impact people’s day-to-day lives. Additionally, there is an extra precaution against harmful short-sighted decisions with long-term negative impacts.